Getting cited by AI search is a chain-position problem before it is a quality problem. You can run every move on the standard checklist for how to get cited by AI search, the schema, the cleaner answer sentence, the extra mentions, and still watch the citation land on someone else's page. Those moves change how your page reads. Position is decided earlier, mostly before you start.
Ahrefs traced ChatGPT's 1,000 most-cited pages and found 67% of them in categories ordinary outreach cannot touch: Wikipedia, homepages, educational domains, app stores. That is the ceiling. The narrow band left under it is the page that holds a number first.
We traced 1,469 blog citations to the end of their reference trail. Only 15.0% ever reached a primary source. That number reframed the question for me: the slot where a number originates sits open far more often than it is filled, and almost nobody is competing for it. The moves below are ranked by how much each one changes your position, starting with the number on the page you care about most.
Citation Goes to the Page the Chain Stops At
The page that gets cited is the one a reference trail can reach and read at the end of a short walk. Schema helps an engine parse you. A clean answer sentence helps it lift you. A strong byline helps it trust you. All three work on a page that already sits somewhere in the chain of who said what.
Position comes first. You set it the day you publish.
This is why the checklist genre keeps letting down the people who follow it. It polishes the wrapper around a number while the citation is decided by the number underneath. The slot you have been optimizing toward is mostly taken, held by reference sites and brand homepages that no amount of formatting moves you into. The one almost nobody is fighting for is the slot where a number starts.
What It Means to Get Cited by AI Search
Getting cited by AI search means an answer engine resolves a claim to your page as the place a number originates. The engine walks the reference chain behind a figure and credits the page where that chain stops.
The walk is short, so walk one number through it. A figure you measured yourself sits in a clean sentence with no link pointing onward, because there is nowhere onward to point. The engine arrives, finds the number stated as its own, and stops. The chain ends at you. Build the same claim on a borrowed figure and the engine takes one more step, to whoever you cited, and the credit settles there, on their page. Across the corpus we traced, the median chain ran about 1.08 hops deep. A trail that short rarely reaches the page that first measured the number.
The deeper questions here have their own homes: why an engine resolves a claim one step short of you, the provenance study behind these numbers, the wider vocabulary of answer engine optimization, and what happens downstream once an agent acts on a number it cannot verify. What matters for the moves ahead is simpler. You want your page at the end of that walk, and everything below is how you get there.
Claim: Only 15.0% of traced SaaS blog citations reach a primary source; the median chain is about one hop deep. Source: LiquiChart 45-domain citation provenance study, 1,469 traced citations. Verified: 2026-06-14.
Own at Least One First-Party Number Per Page
In the corpus we traced, 73% of blog claims were borrowed, the measurement done by someone else and relayed through the page. The post you care about most is unlikely to be the exception. A borrowed number makes the page a hop, one step short of where the chain ends. Measure the number yourself and the page becomes the terminus, with nowhere left for the chain to go.
So the highest-return edit available works on the number itself: convert one borrowed figure on one page that matters into a number you produced yourself. A poll your own readers answer is original measurement. A chart built from numbers you already hold is original measurement. Either one moves that figure from a hop you handed off to a terminus that ends at you.
Picking which number is the whole move.
Not every borrowed figure on a page is load-bearing. When I go through one of our own posts, I am hunting for one number: the stat in the headline, the number a reader quotes back, the figure another blog links when it cites you. That is the one to originate, because that is the one an engine resolves when it credits the page. Convert a throwaway percentage in paragraph nine and nothing about your position changes.
Start with the cheapest source: your own readers. A poll on the question your post already answers turns a borrowed benchmark into a figure you measured, with a sample size you can state and a date you control. After that, the data you already hold: support tickets, usage logs, a survey you ran last quarter. None of it needs a research budget. All of it originates on your page, which is the one property that ends a chain.
Stop reading about other publishers for a moment and count your own most important post. Of the numbers that carry it, how many did you measure, and how many did you inherit from someone else.
The ratio of measured to borrowed numbers on a page decides its chain position before a single tactic is applied, and it is a ratio almost no one has stopped to total. The count is the diagnostic. A page built entirely on borrowed figures is a hop by construction, no matter how clean the formatting, because the chain always runs through it toward whoever it cited.
Before you can convert a borrowed number, you have to know which of your numbers are borrowed, and nothing in the publishing workflow draws that line for you. This is what the claim layer reads: each statistic's position in the chain, so a number you measured surfaces as a terminus and a borrowed one as a hop running through your page toward someone else.
Position decides who gets named when an engine resolves the claim, no matter how the sentence reads. Once you can see the split, the work narrows to one move: take the one borrowed number your page leans on hardest, and make it yours.
Shorten Your Own Citation Chain to the Primary
Some numbers you cannot originate. You did not run the macroeconomic survey or the industry benchmark, and you should not pretend you did. For those, the move is to shorten the chain you sit on, so your page lands one hop from the source rather than adding a link to it.
The rule is small and almost nobody follows it: when you borrow a stat, link the publisher that produced it rather than the blog that re-stated it. That keeps the trail behind your page running straight to the origin. The 15.0% figure above has its own public claim page that resolves hop by hop to where the number started, which is the receipt for what this asks of you. Citation Provenance shows the same depth reading on your own pages, so you can tell a clean one-hop citation from a chain you accidentally lengthened. And before a borrowed figure goes in at all, confirm the number still holds at the other end of the link. A link can resolve perfectly and still point at a figure that has since moved. The full version of this, tracing a claim to its primary source, goes deeper than any single page needs.
Make the Number Reachable and Quotable
A number an engine cannot reach or read cleanly cannot end a chain at you. That makes reachability a position requirement, the kind you meet before the page can be cited at all, which is why it sits here, below owning the number, rather than at the top of the list where the checklists put it. It decides whether a terminus you already hold can actually be read.
Two findings set the priorities. 88% of the URLs ChatGPT cites come straight from search, so a page that has dropped out of classic search has dropped out of the citation pool with it. And cited pages skew about 1.3 years old at the median, so a recent publish date is not what carries a citation. The practical moves are small. State the number in a clean, self-contained sentence an engine can lift on its own. Give the page a plain natural-language URL. Keep it retrievable in ordinary search. In that same study, pages with plain URLs were cited more often than pages with opaque ones, so even the slug is part of whether the number can be reached.
Reachability runs the other way too. Roughly a quarter of the external links in the corpus we traced were dead, gated, or broken, so a chain that points at your number through a failed link never arrives. The figure can be exact and still unreadable.
It is worth being honest about the ceiling here. You cannot out-Wikipedia Wikipedia. Two-thirds of the most-cited pages live in categories you will never occupy through better formatting, which is why reachability earns its keep only in service of the one slot you can take: the page where a number originates.
Keep the Number True Over Time
A number you measured is not finished the day you publish it. Your own data drifts. A figure that was accurate at publish can resolve to a wrong value months later if no one is watching.
The failure is specific. You measured 34% in March. An engine read your page and started citing 34% as the answer in your niche. By autumn your own re-measurement reads 41%, and the engine is still repeating the old number with your name on it, sourced to the page you were proud to own. A terminus you stopped watching is worse than a hop, because the credit is yours and so is the error.
Owning the number is the action. Keeping it true is the discipline. It is worth instrumenting, so you learn a figure has moved before an engine reads the stale one. How a chain gets watched after you publish, and what to do when a terminus shifts under you, is its own piece.
The Citation Was Decided the Day You Published
Citation is a position, and the position was set the day you published, the moment you decided whether the numbers on the page were yours or borrowed. Roughly six in seven chains never reach a primary source, which means the terminus slot is open on almost every page in your niche, waiting for whoever holds the number first. Tracing one page is the free read. Tracing a whole library is what the Citation Scanner is for.
If I had to compress everything above into one instruction, it would be this: own one number per page, the one doing the most work, and keep it true. Everything else, the schema and the reachability and the maintenance, is in service of that one decision.
The terminus slot on your most important post is open right now, behind whatever borrowed number carries it. Measure that one number yourself and the slot is yours. Leave it borrowed and it belongs to whoever you cited.